LineShine: Online Acceleration for HPC-AI Converged Scientific Computing

Abstract

Modern scientific workflows increasingly interleave high-precision numerical simulation, low-precision AI computation, irregular computation, complex communication, and data-intensive processing within continuous execution pipelines. When tightly coupled stages repeatedly cross separate host and accelerator domains, data transfers, explicit synchronization, and distinct software paths can introduce substantial coordination overhead. We present LineShine, a newly deployed exascale supercomputer whose online acceleration architecture provides a new design point for converged HPC and AI. LineShine adopts a many-core online acceleration architecture in which matrix engines are embedded within CPU cores and share a unified execution environment, memory hierarchy, and software stack with scalar and vector execution. This capability is extended to system scale through balanced full-stack co-design, including HBM–DDR hierarchical memory, the Lingqi interconnect, symmetric distributed tiered storage, and an architecture-aware software stack. LineShine achieves leading results on major HPC benchmarks. In the June 2026 rankings, it sustains 2.1984 EFlops on HPL with 80.37% computation efficiency and 52.07 GFlops/W system energy efficiency, and delivers 22.0049 PFlops on HPCG, ranking first on both benchmarks. Beyond benchmark performance, LineShine has supported numerous full-system-scale production applications across a broad range of scientific domains, including computational fluid dynamics, neuroscience, biomedicine, and Earth observation. These applications demonstrate that its balanced architecture and software stack can translate system capability into high throughput, robust scalability, and sustained scientific productivity at exascale.

Keywords

LineShine Supercomputer / Exascale Supercomputing / Online Acceleration / Hardware and Software Co-design / HPC-AI Converged System

Rights & permissions

Higher Education Press 2026

Just Accepted

This article has successfully passed peer review and final editorial review, and will soon enter typesetting, proofreading and other publishing processes. The currently displayed version is the accepted final manuscript. The officially published version will be updated with format, DOI and citation information upon launch. We recommend that you pay attention to subsequent journal notifications and preferentially cite the officially published version. Thank you for your support and cooperation.